Single domain with RAID1+0 disk shelves

[Allen Huang_1]

We have a senario where a DL380 G6 has a P411 disk controller connecting to two external D2700-G6 disk shelfs. Can anyone confirm that P411+D2700 supports single domain and RAID1+0 across the two external shelves in such a way that the data is available even after one external shelf goes down. Is this true?

[Allen Huang_1]

HP tech confirmed that RAID 10 is supported using 2 shelves. But he was not willing to state if the drives from one shelf in a RAID 1+0 set are striped and then mirrored to the other shelf one-for-one. He seems did not want to commit to that if he had not done it himself.

I'm trying to confirm that the RAID 1+0 will stripe â halfâ the disks from shelf one and the other â halfâ the disks on shelf 2, which would be the most redundant possible configuration, however I have not found a document yet that states that is how the RAID 1+0 will work.

Any opinions? Is there a document somewhere states this directly?

[TTr · Accepted solution]

You probably have to try it out and see for yourself. I don't even remember if in the ACU (or ADU report) you can see the pairs within a raid10 array group so that you are able to tell where the disks in each pair come from.

There is a way to force the raid controller to do what you want but it requires a lot of time and patience.

You install one disk in each D2700 and create a raid1 pair. (all other disks are removed). Then you install two more disks (one disk in each enclosure) and add them to the first pair (perform raid type migration from raid1 to raid10). Then keep inserting two more disks and expand the existing raid10. It is doable but as I said it will take a lot of time.

If you are going to put all 50 disks in one raid10 group, the time it will take to do the above manual process is probably prohibitive but if you are going to create several arrays it may be doable.

[gregersenj]

Well, they mirror - then stripe the mirrors.

I also believe it will mirror accross the boxes.
But I'm not 100% sure.
I did test it on a MDS600 a few months ago.
Setup was a C7000 enclosure, 2 SAS Switches and a MDS600 with 2 x 5 drives, connected to both SAS switches.
Only 1 BL460c wit a Smart Array P700m, used the MDS.
It was configured at a RAID 1+0 using all 10 disks.
I testet it unintensionally , but successfully.

I also recommend to test. But to save time, I would go for a "full blow" test
Put in all disks.
Using the ACU: Create an Array using all disks.
Using the ACU: Create a logical drive in RAID 1+0, at the size of e.i 100 GB.
Start formatting from the O/S.
Pull the power from the lower shelf.
If it survive, it's spread over the boxes.

Note:
if a Smart Array suffer from multible disk failure, it will stp using the logical drive(S), that is affected, and disable them.
If you cane recover the drives, then it is possible to re-enable the drives.
From the ACU, or reboot the server, and push the appropriate F-key.

BR
/jag

[Allen Huang_1]

Thank you all for your great suggestions and sharing of your experience. We have tested a configuration where each shelf is separately cabled to each port. So Port1 on P411 goes to one shelf and Port 2 goes to another shelf. We then configured RAID 1+0 with identical disks across the 2 shelves. That worked fine. We unplugged one cable and still saw all the data went across both shelves. When we plugged in the shelf again, it took more than 2 hours to re-synch the mirrors. We set up 2 disk groups (one of each type) that are mirrored to the exact same disks in the other shelf.

[gregersenj]

Yes rebuilding take a long time.
The Smart Array rebuild the disks one at the time. Starting from the first disk.

And you're welcome.

BR
/jag